/*
 * shstr.C
 */

#include <cstring>
#include <cstdlib>

#include <tr1/unordered_set>

#include "shstr.h"

struct hash
{
  std::size_t operator ()(const char *s) const
  {
    unsigned long hash = 0;
    unsigned int i = 0;

    /* use the one-at-a-time hash function, which supposedly is
     * better than the djb2-like one used by perl5.005, but
     * certainly is better then the bug used here before.
     * see http://burtleburtle.net/bob/hash/doobs.html
     */
    while (*s)
      {
        hash += *s++;
        hash += hash << 10;
        hash ^= hash >>  6;
      }

    hash += hash <<  3;
    hash ^= hash >> 11;
    hash += hash << 15;

    return hash;
  }
};

struct equal
{
  bool operator ()(const char *a, const char *b) const
  {
    return !strcmp (a, b);
  }
};

typedef std::tr1::unordered_set<const char *, hash, equal> HT;

static HT ht;

const char *
shstr::find (const char *s)
{
  HT::iterator i = ht.find (s);

  return i != ht.end ()
    ? (char *)*i
    : 0;
}

const char *
shstr::intern (const char *s)
{
  HT::iterator i = ht.find (s);

  if (i != ht.end ())
    return (char *)*i;

  int len = strlen (s);

  int *v = (int *)malloc (sizeof (int) * 2 + len + 1);

  v [0] = len;
  v [1] = 0;

  v += 2;

  memcpy (v, s, len + 1);

  ht.insert ((char *)v);

  return (char *)v;
}

// TODO: periodically test refcounts == 0 for a few strings (e.g. one hash bucket,
// exploiting the fatc that iterators stay valid for unordered_set).
void
shstr::gc ()
{
}

/* buf_overflow() - we don't want to exceed the buffer size of 
 * buf1 by adding on buf2! Returns true if overflow will occur.
 */

int 
buf_overflow (const char *buf1, const char *buf2, int bufsize)
{
    int len1 = 0, len2 = 0;

    if (buf1)
        len1 = strlen (buf1);
    if (buf2)
        len2 = strlen (buf2);
    if ((len1 + len2) >= bufsize)
        return 1;
    return 0;
}